Adverse drug events (ADEs) result in substantial patient morbidity and lead to over 100,000 deaths yearly. The timely identification of previously unknown toxicities of cancer drugs is an important, unsolved problem. In the United States, 20% of the 548 drugs introduced into the market between 1975 and 1999 were either withdrawn or acquired a new black box warning during the 25-year period following initial approval by the Food and Drug Administration. Adverse drug events are an important cause of morbidity and mortality in patients, yet 95% of ADEs are unreported, leading to delays in the detection of previously unknown ADEs and underestimation of the risk to known ADEs. It is known that Electronic Medical Record (EMR), discharge summaries, and lab results contain ADE information and biomedical natural language processing (BioNLP) provides automated tools that facilitate chart review and thus improve patient surveillance and post-marketing pharmacovigilance. The objectives for this proposal are to develop intelligent BioNLP approaches to extract disease, medication, and structured ADE information from EMRs, and then evaluate extracted ADEs for detecting known ADE types as well as clinically unrecognized or novel ADEs whose pattern or effect have not been previously identified.